This blog post describes the paper "Patches Are All You Need?" (Under review, 2021), which was submitted to ICLR2022 (under review as of the end of the Oct.). The ConvMixer proposed in this paper is composed of CNN BN, and unlike previous Vision Transformer series, it can achieve results even on small datasets such as CIFAR. We will then discuss whether patches are really the only important thing with the point of view that the model contains the global information and local information processing mechanisms. In this article, I will explain it according to the following items. The summary of this paper is as follows.

It is put to us on that first Monday morning as an exciting innovation which will streamline our processes and free up time for the important stuff. This happens in our morning "huddle cuddle", where at 9.15am on the dot our manager has us gather around in a loose circle and run through the day ahead – how many pieces of work there are to be processed, which queues will be prioritised, who is going to take lunch when. This is the kind of place where our team of nine refers to each other as family. The kind of place with an A4 printout stuck to the kitchenette fridge that says "if Britney Spears can get through 2007, you can get through today". The job is a data-processing role at a small member-owned health insurance fund.